Mistral Moderation 2
Mistral Moderation 2 is the hosted moderation release from March 2026. It supports a 128k context and includes jailbreak detection. Mistral positions it for multilingual material and long conversations with many turns.
Check the conversation as a sequence
The current model page emphasizes complex multilingual input and long multi-turn conversations. It documents an API service. It does not publish weight access or establish that the model has Shieldstral's policy-question interface.Mistral Moderation 2 API model page
Review errors in conversational context
Practical guidance. Evaluate cases in which a harmless sentence becomes a concern only after earlier turns. Also include quotations and discussion that the product permits. Count missed violations and incorrect blocks separately. A large input limit helps retain context, but it does not establish that every important earlier detail affects the decision.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API |
| Model identifiers | mistral-moderation-2603 |
| Context window | 128,000 tokens The model documentation states 128k context. Mistral Moderation 2 API model page |
| Initial release stage | General availability.Mistral Moderation 2 API model page |
| API identifier | mistral-moderation-2603Mistral Moderation 2 API model page |
| Endpoint | /v1/moderations; batch support is listedMistral Moderation 2 API model page |
Related models
- Shieldstral 1.0
Mistral AI ยท Shieldstral
A policy classifier that checks text and images against plain-language rules and returns yes-or-no scores.
Launch date PreviewDownloadable